Method and system for extracting alphanumeric content from noisy image data
In general, embodiments of the technology relate to extracting content from documents. More specifically, embodiments of the technology relate to using fuzzy regular expressions to process content results obtained from one or more documents in order to extract content for these documents. Further, embodiments of the technology enable the format modification of the content after the content has been identified and extracted from the documents.
1. A method for processing documents, comprising:
obtaining optical character recognition (OCR) results for a document;
processing the OCR results using a matching graph to obtain matching results, comprising:
generating an observation graph using the OCR results; and
determining a best path in the matching graph, the matching graph comprising:
an initial state,
a final state,
at least one matching subgraph node between the initial state and the final state, and
a trash state, between the initial state and the final state, wherein the trash state and the at least one matching subgraph node are not directly connected by any edge in the matching graph,
wherein determining the best path in the matching graph comprises traversing the matching graph at least once, from the initial state to the final state via the at least one matching subgraph nodes, using observation nodes in the observation graph; and
providing the matching results to a user computing system.
2. The method of claim 1 , wherein the observation graph comprises at least one branch.
3. The method of claim 1 , wherein the at least one matching subgraph node comprises an exact match state and a wildcard state.
4. The method of claim 1 , further comprising:
generating the matching graph using a fuzzy regular expression.
5. The method of claim 4 , wherein the matching graph comprises a matching subgraph node for each component of the fuzzy regular expression.
6. The method of claim 1 , wherein the OCR results comprise a set of hypotheses, wherein each hypothesis in the set of hypotheses is associated with a confidence score.
7. The method of claim 1 , wherein the matching graph comprises a plurality of transitions, wherein each of the plurality of transitions is associated with a weight, and wherein the weight of at least one of the plurality of transitions is one selected from a group consisting of a weight for a correct match, a weight for an insertion, a weight for a substitution, and a weight for a deletion.
8. The method of claim 1 , wherein the matching graph comprises a plurality of transitions, wherein each of the plurality of transitions is associated with a weight determined using at least one selected from a group consisting of a matching threshold and a trash score.
9. The method of claim 1 , wherein the matching results comprise at least a partial match of a fuzzy regular expression, wherein providing the matching results to the user computing system comprises modifying at least the partial match prior to obtain a modified match result, wherein the modified match result is provided to the user computing system.
10. A system, comprising:
a processor comprising an integrated circuit;
an optical character recognition (OCR) component executing on the processor, configured to:
obtain OCR results for a document; and
a fuzzy matching component executing on the processor, configured to:
receive the OCR results for the document; and
process the OCR results using a matching graph to obtain matching results, the processing the OCR results comprising:
generating an observation graph using the OCR results; and
determining a best path in the matching graph, the matching graph comprising:
an initial state,
a final state,
at least one matching subgraph node between the initial state and the final state, and
a trash state, between the initial state and the final state, wherein the trash state and the at least one matching subgraph node are not directly connected by any edge in the matching graph,
wherein determining the best path in the matching graph comprises traversing the matching graph at least once, from the initial state to the final state via the at least one matching subgraph nodes, using observation nodes in the observation graph;
wherein the matching results are provided to a user computing system operatively connected to the system.
11. The system of claim 10 , further comprising:
a format modification component executing on the processor, configured to modify at least a partial match of a fuzzy regular expression to obtain a modified match result, wherein the modified match result is provided to the user computing system.
12. The system of claim 10 , wherein the observation graph comprises at least one branch.
13. The system of claim 10 , wherein the at least one matching subgraph node comprises an exact match state and a wild card state.
14. The system of claim 10 , wherein the fuzzy matching component is further configured to:
generate the matching graph using a fuzzy regular expression, wherein the matching graph comprises a matching subgraph node for each component of the fuzzy regular expression.
15. The system of claim 10 , wherein the matching graph comprises a plurality of transitions, wherein each of the plurality of transitions is associated with a weight, and wherein the weight of at least one of the plurality of transitions is one selected from a group consisting of a weight for a correct match, a weight for an insertion, a weight for a substitution, and a weight for a deletion.
16. The system of claim 10 , wherein the matching graph comprises a plurality of transitions, wherein each of the plurality of transitions is associated with a weight determined using at least one selected from a group consisting of a matching threshold and a trash score.
17. A non-transitory computer readable medium comprising computer readable program code, which when executed by a computer processor enables the computer processor to:
obtain a data model for content of a document;
generate a matching graph, the matching graph comprising:
an initial state,
a final state,
at least one matching subgraph node between the initial state and the final state, and
a trash state, between the initial state and the final state, wherein the trash state and the at least one matching subgraph node are not directly connected by any edge in the matching graph,
wherein generating the matching graph comprises using a fuzzy regular expression, wherein the matching graph comprises a matching subgraph node for each component of the fuzzy regular expression and a trash state;
process the data model using the matching graph to obtain matching results; and
provide the matching results to a user computing system,
wherein processing the data model comprises:
generating an observation graph using the data model; and
determining a best path in the matching graph using observation nodes in the observation graph to transition between states of the matching graph,
wherein determining the best path in the matching graph comprises traversing the matching graph at least once, from the initial state to the final state via the at least one matching subgraph nodes, using observation nodes in the observation graph,
wherein the at least one matching subgraph node comprises an exact match state and a wildcard state,
wherein the matching graph comprises a plurality of transitions, wherein each of the plurality of transitions is associated with a weight determined using at least one selected from a group consisting of a matching threshold and a trash score.